J. Child Long. 16 (1989),1-17.Printed in Great Britain

A crosslinguistic investigation of vowel formants in babbling* BENEDICTE D E BOYSSON-BARDIES PIERRE HALLE LAURENT SAGART AND

CATHERINE DURAND

C.N.R.S. & E.H.E.S.S. Université Paris V (Received

2

March 1987. Revised 17 February 1988)

ABSTRACT

A cross-cultural investigation of the influence of target-language in babbling was carried out. 1047 vowels produced by twenty 10-monthold infants from Parisian French, London English, Hong Kong Cantonese and Algiers Arabic language backgrounds were recorded in the cities of origin and spectrally analysed. FI-F2 plots of these vowels were obtained for each subject and each language group. Statistical analyses provide evidence of differences between infants across language backgrounds. These differences parallel those found in adult speech in the corresponding languages. Implications of an early build-up of targetlanguage-oriented production skills are discussed.

INTRODUCTION

T h e term 'babbling' denotes a specific form of production which appears between the ages of six and eight months for the majority of normally developing infants. Infants begin to babble when they utter sounds which exhibit acoustic timing constraints whose characteristics are close to those of mature speech (see Oller 1986 for the requirements that allow the label 'syllable' to be attached to infant production). T h e first patterning of babbling is mainly a reduplication of CV syllables. At around nine to ten

[*I

Pan of this data was presented at the Fourth International Congress for the Study of Child Language and at the Eleventh International Congress of Phonetic Sciences. This research was supported by A.T.P. no. 955138 C.N.R.S. We would like to thank Saleha Chabani and D. Ben Sultana from Algiers, and J. Philips from the MRC Cognitive Development Unit, London, for their help. We are also grateful to the infants' parents who allowed us to conduct our recordings. Address for correspondence: B. de BoyssonBardies, Laboratoire de Psychologie Expérimentale, 54 Bld Raspail, 75006 Paris, France.

CHILD LANGUAGB months, syllabic combinations become more elaborated. Adults cannot recognize words or protowords in babbling productions. What processes are involved in babbling? One view is that babbling is simply the natural output of an immature production apparatus, with no link to perceptual mechanisms. Another view is that perceptuo-motor

attunements are already operating in babbling.

.

The first position can be referred to as the' independence hypothesis '. It postulates that prelinguistic productions are constrained by universal maturational (physiological, biological) processes (Lenneberg 1967) and are thus universal. That is, they do not depend on the infant's linguistic environment. These constraints result in 'phonetic proclivities' (Locke 1983) or 'articulatory proclivities' (Lindblom 1984) which make infants utter subsets of aduit-like productions. According to the independence hypothesis, motor and perceptual components of a language are considered to develop separately (Studdert-Kennedy 1986). Perceptual discrimination skills, already present in neonates, change as a ~sult of the infants' exposure to the phonetic, phonological and intonational characteristics of their future mother-tongue. This leads to some early language-specific discrimination ability (Werker & Tees 1984), not apparent in babbling and first language productions. Although this does not necessarily imply that the linguistic environment cannot affect babbling, such a conclusion is generally drawn. Inter-infant variability of babbling productions within the same linguistic community has been demonstrated both in sound repertoires and in the acoustic characteristics of productions (Ferguson 1979, 1986, Lieberman 1980, Vihman, Ferguson & Elbert 1986). Since inter-infant variability is ascribed to random variations depending on inherent factors, it is regarded as contradicting neither the claim that perception and production in infants are separate processes (Studdert-Kennedy 1986), nor the prediction of universality in babbling productions that follows from that claim. The second position, 'interactional hypothesis', assumes that perceptuomotor mechanisms begin to operate at the babbling stage. According to this hypothesis articulatory procedures that are mastered step by step are oriented by auditory configurations. At 10 months, articulatory control (Buhr 1980, Lieberman 1980, Kent & Murray 1982), together with the restructuring of the auditory system that takes place at around the same age (Werker & Tees 1984), allows infants to specify some vocal tract positions and permits them to produce language-oriented sounds (de Boysson-Bardies, Sagart, Hailé & Durand 1986, McCune & Vihman 1987). Studies of the influence of a target language on babbling are few (N akazima 1962, Weir 1966, Atkinson, MacWhinney & Stoel 1970, Olney & Scholnick 1976, Tuaycharoen 1979, aller & Eilers 1982, Locke 1983, de BoyssonBardies, Sagart & Durand 1984, Thevenin, Eilers, aller & Lavoie 1985, de Boysson-Bardies et al. 1986) and draw conflicting conclusions. Some studies

VOWELS lN BABBLING have found evidence of different patterns of babbling according to the target language. Listeners have been shawn to be able to discriminate babbling of infants from different linguistic backgrounds (Weir 1966, de BoyssonBardies et al. 1984, Ichijima 1987). Also, similarities in long-term spectral characteristic of adults and 10-month-old infants were found for French, Arabic and Cantonese (de Boysson-Bardies et al. 1986). On the other band negative results are reported by Olney & Scholnick (1976) and Atkinson et al. (1970). Prompted by the dominant independence hypothesis, most of the studies in the area have concentrated on the search for universal invariants in babbling through segment counts based on IP A transcriptions. Comparative studies have mainly been limited to frequency counts of consonants (for a comprehensive review, see Locke 1983). However, since the onset of babbling is marked by the appearance of syllabic productions with varied combinations of vowels and consonants, consonant frequency counts alone are not likely to be a good bas~ for crosslinguistic comparisons. Rather, such comparisons should be conducted through the close examination of both consonants and vowels. Phonotactic as weil as the intonational and rhythmic patterns would also constitute necessary additional data for crosslinguistic comparisons of prespeech. A cross-cultural study of babbling from infants reared in English and Spanish-speaking environments in Miami (aller & Eilers 1982.)showed that vowels belonging specifically to the sound inventory of Spanish are generally more frequent with children from the Spanish group while English vowels are on the whole more frequent with children from the English group. ln particular, the vowel [œ] occurred with a frequency of 2.2.'4~~ in English children, but only 13'5 ~,~in Spanish children. However, because of intersubject variability, aller & Eilers did not find any statistically significant differences in the ove rail sound inventories for the two groups. During the past few years various authors have investigated the acoustic characteristics of vowel productions in the first year of life for infants from a single linguistic background. Their investigations have for the most part been restricted to children reared in English-speaking communities. ln particular, the range of variation and average frequencies of the first three formants for 3-, 6-, 9- and 13-month-old infants from an American English background are available (Kent & Murray 1982., Kent & Bauer 1985). Lieberman's( 1980) perceptual and acoustic investigation of the development of vowels in children 16 to 180 weeks old documented the graduaI emergence of a vocalic space in early vocali2.ations. His study gives a week-byweek account of the evolution of average formant frequencies of different yowel sounds, although unfortunately only for those vowels which were recognized as English vowels. Lieberman observed that children produce formant frequency patterns' equivalent' to adult ones, scaled to their shorter 3 1-2

4

VOWELS lN BABBLING ~ ~ ...

" ..

~

.

.D--ü E 0 ~ Z

~ ...

~ f ~

~ 0 >

~

~ ...

'Q~ '" ~ ...

IJ

"0 ~

~ ~ .Q .~ '" ... ~ ~ ... ...

~

'i =

!

ü

~ Q ..

.= '"

'Q-

.5 z ...

... ~ .Q E ~ ;: ~

'§; ~ ~ ~

:if ... ... ~ '" ~ .E ~ ~ ":S c

...

~ ~ Q '" '" ~

~ .~ ~ .Q .-= ... ...

Ij

= ..

~ 0 > '0

os ~ c .. U

c .9 t: 0 Q. 0. .. Co

"0 ;;: G

~ cc

i

~ oS

ë e ~

~

~

-& :ë

.-

"cS

œ c: > >- tI ~ "'=... .. F; "œ c:.. c: œ ~ï: ~ ï: .= = E .~ =' ':ij .:!:= ~ E >.'ë.= ~ c: ~ § œ ~ü~~< ojo~.=,oo~~

2500

..

2

Frequency of F 2 (Hz)

Fig. J. F I-Fz plot of the' mean ,"o"ers' of each infant; infants' linguistic group indicated by terrer.

TABLE

4. Intra-language and inter-language distances under the independencehypothesis (Ho)

English French

English

French

Algerian

0-]8

0-83 0-3'

'"99 0"83 0"42

Aigerian Cantonese :Vlean intra-language :Vlean inter-language

Cantonese z'OO 0,86 0"86 0"..5

distances: 0.393. distances: 1.331.

If we assume that there are no significant differences between infants from different language backgrounds (Ho), distances between infants' vowel sets can be computed by means of a single Mahalannobis distance derived from this population (see Mardia (1977> and de Boysson-Bardies, Sagart & Hailé (1987); see also Appendix). Table 4 summarizes the results under the independence hypothesis (Ho). Infants differ more between different linguistic communities (inter-linguistic community distances) than within any single linguistic community (intralinguistic community distances) as indicated by a Student test (t = z.76, P < 0'05). It is this which allows Ho to be rejected. The independence hypothesis having been rejected, it is also of interest to test the hypothesis which predicts that there is one sub-population per 10

VOWELS lN BABBLING TABLE

S. Intra-language and inter-languagedistancesunder interactional hypothesis(HI)

English French Alj!erian Cantonese

o'.J6

0-..7 0-29

.'36 .'00 0-.5 0,60

:\Iean intra-languagc distances: 0""'°. :\Iean inter-langua~e distances: 0'767-

linguistic community (H 1 hypothesis). ln this way. we should get a more accurate estimation of the distances between inter-linguistic communities. A number of different Mahalannobis distances are now involved in the calculation to test HI. Table 5 summarizes the results: averaged inter-linguistic community distances are larger th an avera~ intra-linguistic community distances. To summarize. differences in vowels across linguistic communities already emerge in lo-month-old infants. ln addition the interaction hypothesis yields an estimation of inter-linguistic community distances which might be useful when examining the second issue: is there a similarity with adult vowels ?

Reflection on adult vowels The main tendencies of infant vowels can be roughly visualized by means of what we called .mean vowels', defined per infant or per language community. The F2/FI ratio, an index of vowel compactness, is also informative in this respect, as will be seen below. ln this section we attempt to give a similar account of mean vowels and F2/FI ratios for adult speakers in the four languages under scrutiny. We used two kinds of data from the existing literature on each language: frequency counts of vowels computed out of large corpora of running speech, and FI and F2 mean values for individual vowels, found in exhaustive acoustico-phonetic studies involving a number of speakers. For each language, FI (respectively F2) of the mean vowel was computed as the frequency-count-weighted mean of FI (respectively F2) values from the acoustico-phonetic study. For English (R.P .), French and Hong Kong Cantonese, we used frequency counts from Fry (1947), Wioland (1972) and Fok (1979) respectively. For formant values we used Henton (1983), Longchamp (in press) and Lee (1985) respectively. Data on Algiers Arabic could not be found, 50 instead we used data on Moroccan Arabic whose vocalic system is rather close to that of

CHILD

LANGUAGE

Aigiers dialect. Both frequency counts and formant values for Moroccan were found in Benkirane (1982.). Since nasalized and diphthongized vowels of babbling were not considered, nasalized vowels in French and diphthongs in English and Cantonese were not taken into account in this survey. As for ail four languages, formant values for schwa were not documented, schwa had tobe discarded from the mean vowel computation. FI and Fz mean vowel values computed for English, French, Moroccan and Cantonese are shown in table 6 together with the Fz/FI ratios.

TABLE

6. Mean formant frequencies (Hz) and F2/Fr by language community

FI Fa Fz/FI

Enl:lish

French

Moroccan

459 16~ 3'68

~6S ISZ9 3"z8

506 1534 3'°3

ratios ofadult t.'owels

Cant()nese

536 1457 Z"71

The same trends are found for adu!t speech and babb!ing: Eng!ish and French have more diffuse vowe!s th an Cantonese and Moroccan. The two most different languages in this respect, as with infants, are Eng!ish and Cantonese in adu!t speech as weil as in babb!ing. This cou!d have been predicted from the examination of frequency counts for adu!ts which indicate that English favours vowe!s which are high, front or both ([1], [e] are the most frequent vowe!s not counting schwa) and Cantonese favours !ow, back vowe!s ([0:], [:>:], [8], are the most frequent vowe!s). lndeed, the implications of these trends agree with the fact that Eng!ish has, on average, rather diffuse vowe!s and Cantonese rather compact ones.

DISCUSSION

The results support the fundamental claim of the interactional hypothesis that the variability in Io-month-old infant productions can be inftuenced by the characteristics of the linguistic environment. There is a close similarity between infant and adult vowels across the four linguistic communities examined here. This is best illustrated by the parallel patterns of the FI/Fz ratios found in infant and adult vowels. Such a conclusion challenges the dominant independence hypothesis. 1t reinforces the position that claims that there is a link between auditory and articulatory configurations and that the perceptuomotor system is already functional in the second half of the first year. The early influence of environment shown in our data is not incompatible 13

VOWELS lN BABBLING with trends generally thought of as universals. However, these trends, such as a tendency to centralize vocalic space, or the predominance of front vowels over back vowels, are modified by the characteristics of the vocalic linguistic environment. The ranges of values reported here for FI and F2 are also consistent with observations on infants of similar age from AmericanEnglish linguistic background (Buhr 1980, Lieberman 1980, Kent & Murray 1982): 700-1000 Hz for FI and 2250-3000 Hz for F2. However, when infants from more than one linguistic community are studied together, a wider range of values is found. From our data, inter-individual variability appears to be quite large: the first formant of infants' mean vowels varies from 758 Hz (one French infant) to 1173 Hz (one infant from Hong Kong), whereas the second formant varies from 2061 Hz (one infant from Hong Kong) to 2805 Hz (one English infant). Six infants (two English, two French, one Algerian and one Cantonese) out of the 20 cluster around the average values of FI and F2, within the 900-1000 Hz range for FI a8d the 2400-2600 Hz range for F2. The spreading of infants' vocalic spaces is also quite variable: some infants produce a wider spectrum of vocalic sounds than others. Ali this is in agreement with commonly observed inter-infant variability in babbling productions. Nevertheless, according to our results, the variability related to acoustic characteristics of vocalic space should not be ascribed to individual differences only, since inter-group differences have been found to be significantly larger than intra-group ones. These systematic differences between groups cannot be attributed exclusively to physical differences in infants, such as vocal tract length, which are probably the most influential with respect to formant frequencies. lndeed, if the differences between two given groups were only a matter of vocal tract length, the scaling of formant frequencies would constitute the only difference, with bath FI and F2 yielding higher (or lower) values for a given group. This is far from the case. For example, in the case of the English and Cantonese groups, English infants' vowels have a lower F l, but a higher F2 than those of Hong Kong infants. Clearly, the difference does not lie in the scaling of formant frequencies. Though individual differences in vocal tract length cannot be denied, they do not appear to be systematically related to membership in particular linguistic communities. The systematic differences between infant groups fit better the differences between the average acoustic characteristics of vowels found in the adult languages. FI values for English infants are generally low (the mean FI is 876 Hz, lower th an for the Cantonese and Algerian groups) and only somewhat scattered. Mean F2 is higher than for any other group (2628 Hz). The average values of FI and F2 observed by Kent & Murray (1982) in a group of 9-month-old American infants, about 900-1000 Hz for FI and 3000 Hz for F2. agree better with our data for English infants than with those for any 13

CHILD LANGUAGE other group. The F2/FI ratio is highest for English infants. It is also highest for English adults compared to French, Moroccan and Cantonese speakers. Individual English children do not depan from the general tendency of the English group. as shown in Fig. 2(a) and Fig. 3. ln Fig. 3. aIl the mean vowels for English infants gather in the upper left pan of the formant chart. This illustrates the preference of ail English infants for diffuse vo\vels, in spite of individual variations. FI values for French infants are also low (mean FI = 878 Hz). French mean F2 is between that for English and that for Cantonese. The F2/FI ratio is lower than for English but higher than for Algerian and Cantonese infants. The lame pattern was found in adult speech. As can be seen in FiR. 3. the mean vowels for French infants are closer to those of English infants than to those for the other groupa, ".ith a lower F2, which suggests a tendency towards a less fronted articulation. Aigerian infants have intermediate FI values, and low F2 values, as low as for Cantonese. This paral];ls the preference for the Law Central vowcl [a] in adult speech. The F2/F 1 ratio is lo\ver than for English and French. and higher than for Cantonese. Again, the same pattern was found in adult speech. Infants in the Cantonese group show a preference for high FI and low F2 vowels. Their mean FI is higher than for any other group. and their mean F2 is as low as for the Algerian infants. The F2/FI ratio is lower than for any other group. just as in adult speech. This suggests a marked tendency towards vowel compactness. Two infants depart from this general tendency. one for FI and the other for F2 values. However. the mean vowels in the Cantonese group do not intermix with either the English or the French groups. To summarize, we claim that the functional organization of the aniculatory principles for speech in different languages begins during the babbling stage. ln an earlier study (de Boysson-Bardies ~t al. 1986), we claimed that setting up 100se articulatory limits to tongue and lip movements in vowel production couJd be the first step towards acquiring the vowel system of a target language. This implies that speech sounds in the surrounding language provide material for building the internaI representations that infants use to try out articulatory patterns and configurations of the vocal tract. Werker & Tees (198...) indicate that there is a decline in the discrimination of non-native speech contrasts around 9-10 months and conclude that it is a function of specific language experience. We show that vocalic productions of infants around this age reftect some of the systematic differences in the mode of articulatory positioning that characterizes different languages. The data concerning perception and production tend to agree on the fact that the linguistic environrnent modifies the child's general abilitics at the babbling stage.

VOWELS lN BABBLING According to Locke (1983: 94) language acquisition begins . when a child moves away from what would continue to be his pattern, and cI oser to the ambient one '. ln Locke's view this takes place after the first fifty words have been acquired, when the rate of acquisition of lexical items increases markedly. ln the present study we have shown that the build-up of languageoriented articulatory skills is already emerging at the end of the first year.

APPENDIX Under Ho, aIl the infants' vowel sets are assumed to belong to a single population characterized with a covariance matrix ~.. computed from aIl data. ~.. defines the Mahalannobis distance for estimating distances between vowel sets of any two infants:

~W.JI -1 4 " - 4 -..

.

wherek and1areindexesfor linguisticcommunities, i andj

indexes for infants within their linguistic communities, and .1 the vector of differences between Fi means (i = 1 to 2.). Average intra-linguisric community distances are computed according to (4 ~ 4 6 Dk~1

1: 1: nkl"ki

)/(

d:'.t/

"-1/-1+1

~ 1: nkj

nki

1-1/-/+1

)

(~}

where nkf is the degree of freedom of the vowel sets and Dk the average intralinguistic community distance within the kth linguistic community. Similarly, the averaged inter-linguistic distance between the kth and [th linguistic communities is given by: 6

6

~ ~ n~n'Ill:t'IJ

)/(

6

6

~ ~ n~nlJ

1-1/-1

.1-1/-1

)

(2)

Under HI, infants' vowel sets are grouped by linguistic communityo The kth linguistic community is characterized by the covariance matrix J:to lntra-linguistic community distances are:

~.tJ

- d};;ttl",

while inter-linguistic community distances are obtained by averaging the covariance matrices of the linguistic communities involved:

~.II- .lfI;.\JT. 15

CHILD LANGUAGE where E~.I

-(X~E~+",E,)/(n~+n,)

and nk is the degree of freedom of the entire vowel set of the kth linguistic community. Averaged intra- and inter-linguistic community distances are computed as in (1) and (z). REFERE:"-CES Atkinson, K., MacWhinney, B. & Stoel, C. (1970). An experiment in the recognition of babbling. PsychologicalRn'iew of Child langt4ageDn'elopment 1. 71-6. Bo:nkirane,T. (1982). Etude phonitique et fonctions de la syllabe en arabe marocain. Thèse de troisième cycle, Université d'Aix-en-Provence. Boysson-Bardies, B. do:, Sag-..rt, L. & Durand, C. (1984). Discernible ditferences in tho: babbling of infants according to target-language.Journal of Child languageIl. I-IS. Boyss.m-Bardies, B. de, Sagart, L., Hailé, P. & Durand, C. (1986). Acoustic investigation of cross-linguistic variability in babbling. ln B. Lindblom & R. Zetterstrom (eds), PreCl4rsors ofearly speech.:"-ew York: St~ton Press. Boysson-Bardies, B. de, Sagart, L. & Hailé, P. (1987). The influence of target language on vocalic spacein ten-month-old infants. ln Proceedingsof the Eln'enth International Congress nf Phonetic Sciences.Tallinn, Estonia. Buhr, R. D. (1980). The emergence of vowels in an infant. Journal of Speechand Hearing Rt'search23. 62-. Multivariate analysis. 1\'. ln P. R. Krishnaiah (ed.), "'Iahalallnobis distallcesand angles.New York: t\'orth Holland. McCune, L. & Vihman, M. (1987). Vocal molor schemas. Papers alld Reports on Child LanguageDet'elopment36. 72-9. Nakazima, S. (1962). A comparative study of the speech development of Japanese and American English in childhood. Stltdia Phonologica2. Z7-39. Olier, D. K. (1980). The emergence of th" sounds of speech in infancy. ln G. YeniKomshian, C. Kavanagh & C. Ferguson (eds), Child phonology:pe,ceptioll and production. t\'ew York: Academic Press. (1986). Metaphonology and infant vocalizations. ln B. Lindblom & R. Zetterstrom (eds), P,ecursorsof ea,ly speech.~w York: Stockton Press. Olier, D. K. & Eilers, R. E. (198z). Similarity of babbling in Spanish and English-learning babies. Journal of Child Langllage 9. 565-77. Olney, R, & Scholnick, E. (1976). Adult judgements of ageand linguistic differences in infant vocalisations. Journal of Child Language3. 145-56. Studdert-Kennedy, M. (1986). Sources of variability in early speech development. ln J. S. Perkell & D. H. Klatt (eds), Invariance and t'a,iability in speechp,ocesses.Hillsdale NJ: Erlbaum. Thevenin, D. M., Eilers, R. E., Olier, D. K. & Lavoie, L. (1985). Where is the drift in babbling? A cross-linguistic study. Applied Psychalinguistics6. 1-15. Tuaycharoen, P. (1979). An account of speech development of a Thai child: from babbling to speech. ln T. L. Thongkum, V. Panupong, P. Kullavanijaya & M. R. Tingsabadh (eds), Studiesin Thaï and Mon-Kmer phaneticsand phonology:in honor of EugenieJ. A. Henderson. Bangkok: Chulalongkom University Press. Vihman, M. M., Ferguson, C. A. & Elbert, M. (1986). Phonological development from babbling to speech: common tendencies and individual differences. Applied Psycholinguistics 7. 3-40. Weir, R. W. (1966). Some questions on the child's leaming of phonology. ln F. Smith & G. A. Miller (eds), The genesisof language,Cambridge MA: M.I. T. Press. Werker, J. & Tees, R. C. (198..). Cross-languagespeechperception: evidence for perceptual reorganization during the first year of life. Infant Behavior and Development7. ..9-63. Wioland, F. (1972). Estimation de la fréquence des phonèmes en français parlé. Travaux de

r Institut dePhonétique..177-20".